Distribution Power System Outage Diagnosis based on Root Cause Analysis

Abstract

This paper proposes data mining-based models to diagnose outage data in distribution power systems. In this work, outage data from a local distribution company is gathered and aligned with weather data. Then, a subset of features is selected to reduce the processing time and simplifying purposes. To increase the fairness of final models and to account for differences in misclassification cost, using a customized cost matrix is proposed. Two decision tree-based modeling algorithms are trained and tested. Results show the ability of the established models to diagnose the root cause of an outage fairly well. In addition, an ensemble of the decision tree-based models is built, which outperforms the other two models in almost all cases. Finally, applications of such models in decreasing outage duration and improving the reliability of the power distribution network are discussed.